A/B testing and experimentation is currently common among companies offering services online, particularly social media. Ability to collect various data and related usage metrics from customers makes this experimentation design useful in solving if new implementations or already existing factors are causing the outcome. A possible and disrupting occurrence in this scenario is interference. Han et al. (2022) give a motivating example of an A/B test implemented by LinkedIn in which interference is a likely cause. To solve this issue, they propose a set of algorithms to test for interference in various cases. In this paper algorithms 1-3 from the study are replicated and compared to the original paper. Lastly, some considerations for implementing the algorithms are given.
Replication paper containing the literature review and code of the project in R -
"Project Paper - Interference testing with increasing allocation.pdf"
R code based on the interpretation of the first 3 algorithms described in the paper
"Algorithm code 1-3.R"